On Thu, Mar 23, 2000, Dunlap, Randy <[EMAIL PROTECTED]> wrote:
> [I'm not quoting all of this thread's discussion.
> If you are very interested, you can look in the
> email archives for it.]
>
> Right now we have 3 drivers in the 2.3.99 kernel
> that could/should be in user space, assuming that
> usbdevfs provides all that they need.
> rio500
> dc2xx
> mdc800
>
> I'm not trying to have these removed from the 2.3/2.4
> Linux kernel. This could be the wrong decision,
> but I'd like to see the device support available.
> Period. If I removed them right now, then the
> device support wouldn't be available for some
> unknown period of time.
>
> So I agree with most of Johannes' arguments, such as
> if Java is won't do it, don't use Java (to paraphrase),
> just maybe not his timeframe.
>
> JE wrote:
> > Everything is a tradeoff, but I see no overwhelming reason to keep the
> > driver in the kernel, whereas there are tons of good, tangible reasons
> > to move it to userspace.
>
> I see one good reason to keep these 3 drivers (and maybe scanner ?)
> in the kernel right now: user satisfaction. I'm trying to be
> realistic and pragmatic, not idealistic.
Yes, I agree with that. Something needs to be working, and until
userspace drivers are complete and working satisfactorily, the existing
should stay.
> We still need to see at least one of these done in
> user space.
>
> Bruce Tenison: has the rio500 team done
> anything about this evaluation? Have you
> discovered any problems (like missing functionality
> or bugs) in usbdevfs? or in libusb if you are trying
> to use it?
>
> David Brownell: what features/functionality do you
> think must be added to usbdevfs to make it complete?
>
> Henning Zabel: what is keeping you from making the mdc800
> driver a userspace-only solution?
>
> Please try to answer with detailed suggestions, not vagueness.
>
> I don't know whether usbdevfs provides everything that
> is needed right now for a userspace driver solution.
> My bet is that it will within a few weeks or months even
> if it doesn't right now. Given that proviso, I expect/plan
> to remove these 3 drivers from the Linux 2.5 kernel.
One of the problems brought up by many people (including David Brownell)
is the device permission/ownership problem with userspace drivers.
If a userspace driver wishes to access a device as a non privileged
user, root must chown/chmod the device node, after everytime the device
is plugged in.
I've worked on this part of today (before this medicine I've been
taking for the last couple of days for this virus knocked me out), working
with usbd and devfsd to make this seamless. I'm going to post something
tomorrow. For now, it's a inconvenice for development purposes.
I've also been working on making libusb feature complete. Things like
control messages and bulk transfers are there, but asynchronous
transfers are not. For dc2xx, this shouldn't be a problem. I haven't
looked at the rio500 and mdc800 drivers to see what they do. Some
generic descriptor parsing routines will also be implemented soon.
FWIW, I've already implemented USB support for gphoto (in the 0.5
development tree) and I have a dc2xx driver I've implemented which I
have sitting here but never got around to submit. David may not need to
do anything, unless he wishes to add support to a library such as
opendis or the such.
JE
---------------------------------------------------------------------
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
